If you are trying to minimize the “risk” associated with a dynamically generated finding relating to human behavior, there are many methods. A typical method in my field is to conduct multiple day probes of individuals/ entities in the target categories to form “personas” or “clusters”. This provides you with a data set of typical movements and activities throughout the day. week, month.
Relating those behaviors to specific data streams (ie time stamp / location) can provide a bandwidth for determining if the finding falls within a typical cluster or is an outlier. If an outlier is unique it is probably an error, if it correlates to another “persona” it may not be.
This can similarly apply to projected behaviors.
I am from a different field, not sure if this helps / is obvious.